AnchorLookupOptions.contains: add benchmark + verify flag-on latency < 2x flag-off (cross-repo with ooxml-swift)

[kiki830621]

Problem

From verify of #115 (DA P3 #6):
「Performance：AnchorLookupOptions.contains 對全文做 O(N×M) 掃描，沒 short-circuit、沒 benchmark。1000+ paragraph 文件、多次 lookup 場景未驗證 perf。」
— Source: team:devils-advocate (P3)
Verify report: #115 (comment)

Type

perf / observability

Strategy

1. 加 benchmark target：

// In ooxml-swift Tests
final class AnchorLookupBenchmarks: XCTestCase {
    func testLookupLatency_1000Paragraphs_FlagOff() throws {
        let doc = makeDocxWith(paragraphs: 1000, avgChars: 200)
        measure {
            for _ in 0..<100 {
                _ = doc.findText("typical anchor", options: .exact)
            }
        }
    }

    func testLookupLatency_1000Paragraphs_FlagOn() throws {
        let doc = makeDocxWith(paragraphs: 1000, avgChars: 200)
        measure {
            for _ in 0..<100 {
                _ = doc.findText("typical anchor", options: .mathScriptInsensitive)
            }
        }
    }
}

報告 ratio：flag on 應該 < 2x flag off latency。

2. 若 ratio > 2x，考慮優化：

• Memoize per-paragraph normalized form on demand
• Pre-compute flattenedNormalized field on Paragraph load（trade memory for repeated lookup speed）
• Short-circuit: if needle has no Unicode subscript / superscript chars, fall back to .exact 路徑而非走 normalized

Test

• Benchmark target run on CI、結果記錄為 baseline
• Lib 加 measurement test，PR 不必 ship 結果但要有測試可跑

Related

• Add math-script match options to insertion anchors #115 (PR), P3 enhancement: H₀ Unicode subscript anchors don't match flatten output (post-#85 follow-up) #90
• ooxml-swift cross-repo（lib-level perf）